If the problem is operating a proxy, replace LiteLLM with OpenRouter or Requesty and pay a 5% to 5.5% fee instead. If it is guardrails and governance, Portkey does that as its primary product. If it is dashboards, Helicone. If your platform team already runs Kong, Vercel, or Cloudflare, each has a gateway that costs nothing extra to adopt. And if the traffic is coding agents rather than application requests, none of these is the right shape.
- Nobody wants all five LiteLLM jobs replaced. Identify the one that broke.
- Do not want the ops at all: OpenRouter (5.5% on top-ups) or Requesty (5% on inference).
- Want guardrails and RBAC: Portkey. MIT gateway, $49/mo cloud for 100k logs.
- Want observability: Helicone. Hobby free, Pro $79/mo.
- Already on a platform: Kong, Vercel (0% on tokens), or Cloudflare (free core).
- Running coding agents: a request proxy is the wrong shape entirely.
Which of the five jobs broke?
| What stopped working | Look at |
|---|---|
| Nobody owns patching or upgrades | OpenRouter, Requesty, Portkey Cloud, Helicone Cloud |
| Config sprawl and YAML review fatigue | Portkey, Vercel AI Gateway, Cloudflare |
| Need guardrails, PII redaction, RBAC | Portkey, or LiteLLM Enterprise |
| Need sessions, evals, per-user cost | Helicone |
| Need an API gateway your ops team already runs | Kong AI Gateway |
| Need models you have no accounts for | OpenRouter |
| Need it to cost nothing at all | Vercel (0% on tokens), Cloudflare (free core) |
| It is coding-agent traffic, not app traffic | Continuum, or BYOK direct |
The alternatives
OpenRouter, if you do not want the ops
The cleanest trade: delete the deployment, delete the patch cadence, delete the config repo, and pay 5.5% on credit purchases instead. It brings a catalogue you do not otherwise have (415 model entries across 102 provider endpoints on 19 August 2026) and takes away your ability to keep prompts inside your network. If that second clause is why you chose LiteLLM in the first place, stop here. Detail in what OpenRouter is and the head-to-head.
Requesty, the hosted router with cleaner pricing
A hosted gateway at https://router.requesty.ai/v1 with a flat 5% markup on inference, no subscription, no seat fees, and no minimum spend, advertising 600+ models across 30+ providers. It has cost, latency, and availability routing, failover chains, weighted load balancing, and auto-caching, which is most of what people use LiteLLM's Router for. Enterprise adds SSO, RBAC, guardrails, and PII detection, plus EU data residency. See the review.
Portkey, if governance is the requirement
The most direct like-for-like, because it is also a gateway with an open-source core you can self-host (MIT, 12,766 stars) and a cloud you can buy instead. The difference in emphasis is guardrails: input and output policies, PII controls, and virtual keys are the product rather than an enterprise upsell.
| Plan | Price | Logs | Retention |
|---|---|---|---|
| Open source | Free | No limit (self-hosted) | Yours |
| Developer | Free | 10k logs/month | 3 days logs, 30 days metrics |
| Production | $49/month | 100k logs/month, then $9 per 100k | 30 days logs, 90 days metrics |
| Enterprise | Custom | 10M+ logs/month | Custom, plus VPC and SOC 2 Type 2 |
Helicone, if the real need is visibility
A lot of LiteLLM deployments exist because someone asked what the AI bill was made of. If that is the whole reason, an observability platform answers it better than a router does. Helicone gives sessions, per-user metrics, prompt management, datasets, a playground, alerts, and its own query language, plus a gateway at https://ai-gateway.helicone.ai. Hobby is free with 10,000 requests, Pro is $79/month, Team is $799/month. Read the review first, including the note about its self-hostable gateway repo.
Kong, Vercel, and Cloudflare, if you already live there
The cheapest gateway is one your platform team already operates. Kong AI Gateway extends an API gateway many organizations already run, which means the auth, rate limiting, and observability story is already answered and already reviewed. Vercel AI Gateway charges no markup and no platform fee on tokens, including on BYOK, and meters only optional extras such as $0.10 per 1,000 requests for a team-wide provider allowlist. Cloudflare AI Gateway gives analytics, caching, and rate limiting free, with 100,000 logs on Workers Free and 10,000,000 per gateway on Paid; only its unified billing adds 5%.
Side by side
| Self-host | Fee on tokens | Governance | Observability | |
|---|---|---|---|---|
| LiteLLM | Yes (MIT core) | None | Enterprise licence | Prometheus, callbacks |
| OpenRouter | No | 5.5% on top-ups | Basic | Usage dashboards |
| Requesty | No | 5% on inference | Enterprise tier | P50-P99, cost, cache |
| Portkey | Yes (MIT) | None (log-priced) | Its main product | Logs, traces, metrics |
| Helicone | Yes (Apache 2.0 platform) | None (log-priced) | Rate limits, alerts | Its main product |
| Vercel | No | 0% | Allowlist, ZDR (metered) | Trace drains (OTel) |
| Cloudflare | No | 0%, or 5% on unified billing | DLP, guardrails | Free analytics |
When none of these is the answer
Every product above assumes an application making API calls that should pass through a proxy. A meaningful share of LiteLLM deployments are not that. They exist because an engineering org wanted to know what Claude Code and Codex were costing, per team, and a gateway was the only lever anyone could think of. That is a heavy way to solve it: you end up routing an interactive coding agent through a proxy you now operate, which adds a hop, a failure mode, and a compatibility surface to the most latency-sensitive tool your developers touch.
Questions people ask
What is the best LiteLLM alternative?
Portkey is the closest like-for-like: an open-source gateway with a hosted option and guardrails as a first-class feature. If the point is to stop operating anything, OpenRouter or Requesty. If the point was always the dashboards, Helicone. If your team already runs Kong, Vercel, or Cloudflare, use theirs and add nothing new.
LiteLLM vs Portkey: which should I use?
LiteLLM if you want maximum provider coverage and a free, unrestricted self-hosted proxy, and you have someone to operate it. Portkey if guardrails, PII controls, and RBAC are the requirement, since those are its product rather than an enterprise add-on, and if you would rather buy a cloud plan ($49/month for 100k logs) than run a deployment.
Is there a LiteLLM alternative with no per-token fee?
Several. Portkey and Helicone both self-host and both price on logs rather than tokens. Vercel AI Gateway is hosted and charges no markup and no platform fee on tokens at all, including with your own keys. Cloudflare AI Gateway is free for its core analytics, caching, and rate-limiting features.
Can I replace LiteLLM with OpenRouter?
Only if you can accept a third-party data processor. OpenRouter removes the deployment, the config, and the patch cadence, and gives you a much larger catalogue, for 5.5% on credit purchases. It cannot keep prompts inside your network, which for many LiteLLM users is the whole reason they self-hosted.
What is Bifrost compared to LiteLLM?
Bifrost is one of several newer Go and Rust gateways positioned on raw throughput and lower per-request overhead than a Python proxy. The trade is maturity and integration breadth: LiteLLM has 100+ provider integrations, 56,762 GitHub stars, and an ecosystem that already assumes it. Benchmark on your own traffic before switching for performance, because in most deployments the proxy is not the bottleneck.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- LiteLLM enterprise open source vs enterprise feature split
- Portkey pricing $49/mo Production, 100k logs, $9 per 100k overage
- Portkey-AI/gateway on GitHub MIT licence and star count read 19 Aug 2026
- Helicone pricing Hobby free, Pro $79/mo, Team $799/mo
- Vercel AI Gateway pricing no markup on tokens or BYOK, add-on surcharges
- Cloudflare AI Gateway pricing free core features, log limits, 5% unified billing
- Requesty pricing 5% markup, no seats, no minimum